Accurate modeling and prediction of driving behavior are crucial for enabling autonomous vehicles to safely navigate complex, interactive traffic environments. While recent continual learning approaches for interactive trajectory prediction aim to learn efficiently from streaming data, they often fail to fully retain previously learned cases when acquiring new knowledge, a phenomenon we term case-level forgetting. This limitation poses significant risks in safety-critical autonomous driving applications. This study identifies, analyzes, and addresses case-level forgetting in continual learning for trajectory prediction. We propose the dynamically expandable interactive trajectory predictor (DEITP), a novel framework that preserves previously learned knowledge through a dynamic model expansion mechanism. The mechanism regulates expansion timing by assessing model similarity, thereby controlling model growthwhile preventing catastrophic forgetting. Furthermore, to operate in realistic task-free settings where task identity is unavailable at test time, we introduce a task identification strategy based on a familiarity autoencoder that selects the most appropriate expert for prediction. Extensive experiments on real-world datasets demonstrate that DEITP substantially mitigates forgetting and achieves zero-forgetting performance when task identities are known.

Vehicle lateral control is an important subtask of vehicle autonomous driving. There are many external disturbances that will affect the lateral control accuracy of the vehicle, and the inclination of the road is one of the most important ones. The inclined road will lead to additional lateral forces on the vehicle and will also change the magnitude of support force on the vehicle. The change of lateral force and support force will ultimately affect the trajectory tracking performance of the vehicle. Most of the current trajectory tracking methods only consider the trajectory tracking problem on the plane. If the influence of the road surface is considered in the design of the vehicle's trajectory tracking controller, the dynamic response and the tracking accuracy of the vehicle can be improved. This paper proposes a method based on Udwadia–Kalaba equation to calculate the normal and lateral force on a vehicle tracking a desired trajectory on an inclined road. Further, a trajectory tracking controller that considers the road inclination is designed. Finally, the simulation of trajectory tracking performance with an inclination angle is carried out to verify the effectiveness of the proposed controller.

Railway transportation plays an important role in modern society. As China’s massive railway transportation network continues to grow in total mileage and operation density, the energy consumption of trains becomes a serious concern. For any given route, the geographic characteristics are known a priori, but the parameters (e.g., loading and marshaling) of trains vary from one trip to another. An extensive analysis of the train operation data suggests that the control gear operation of trains is the most important factor that affects the energy consumption. Such an observation determines that the problem of energy-efficient train driving has to be addressed by considering both the geographic information and the trip parameters. However, the problem is difficult to solve due to its high dimension, nonlinearity, complex constraints, and time-varying characteristics. Faced with these difficulties, we propose an energy-efficient train control framework based on a hierarchical ensemble learning approach. Through hierarchical refinement, we learn prediction models of speed and gear. The learned models can be used to derive optimized driving operations under real-time requirements. This study uses random forest and bagging – REPTree as classification algorithm and regression algorithm, respectively. We conduct an extensive study on the potential of bagging, decision trees, random forest, and feature selection to design an effective hierarchical ensemble learning framework. The proposed framework was testified through simulation. The average energy consumption of the proposed method is over 7% lower than that of human drivers.